In medical imaging, image registration (the alignment of two images with each other such that both are placed in a common coordinate system) generally requires the identification of features in each image, such as edges, bright spots and the like. Having identified such features, image registration algorithms attempt to match similar features between the images. However, different images may have widely varying contrasts, colour schemes and the like, complicating feature identification. This may be especially true when attempting to register images taken with different modalities (e.g. MRI and ultrasound).
Current image registration techniques may therefore result in inaccurate registration, particularly in medical applications, and may also impose onerous computational burdens.